SEO and AEO audit: a practical guide for B2B teams

An SEO and AEO audit examines two connected but different systems. The SEO side tests whether search engines can access, interpret, index, and surface useful pages. The AEO side reviews whether your information is clear and well-supported enough to answer buyer questions, then samples how named answer products actually represent and cite that information.

A combined audit is valuable because both disciplines depend on accessible, useful, trustworthy source material. It becomes misleading when variable generated answers are treated like deterministic crawl checks or when a temporary citation is sold as proof of a permanent “AI ranking.”

For a B2B team, the deliverable should be a prioritized backlog with evidence, owners, dependencies, and retests—not a new layer of acronyms.

Set the audit around revenue and buyer decisions

Start with business context. Identify the offers, audiences, markets, competitors, sales motions, and conversion paths that matter. List the pages responsible for category education, product evaluation, proof, implementation, pricing, and conversion.

Then define the questions the audit must answer. For example:

• Can search crawlers reliably access and render our product and solution templates?
• Do our pages match the language and decisions of qualified buyers?
• Are important product and company claims current and consistent?
• Which sources appear for category and comparison prompts?
• Are generated descriptions of our company materially accurate?
• Can we connect discoverability to qualified site actions without overstating attribution?

Agree on domains, subdomains, locales, templates, data access, and exclusions. A bounded audit may not include log analysis, backlink remediation, migration support, implementation, or ongoing monitoring. Explicit boundaries make proposals comparable and findings credible.

Follow the dependency chain from access to outcome

A combined audit should work through five layers in order:

1. Access and rendering: Can approved crawlers fetch the intended page and its important content?
2. Index and retrieval eligibility: Do directives, canonicals, internal links, and sitemaps support the intended URL?
3. Content and evidence: Does the page answer a buyer task with clear, supported, current information?
4. Entity and corroboration: Do organization, product, people, and claim details agree across owned and relevant external sources?
5. Observed visibility and outcome: Do sampled search and answer experiences show the brand, and do identifiable visits create useful business actions?

This order prevents wasted work. Rewriting answer blocks on a non-canonical or inaccessible template is premature. Fixing crawl access on a page with no distinct buyer purpose is also insufficient.

Google says its AI search features use the same foundational SEO requirements as Search, and a page must be indexed and eligible for a snippet to be considered (Google Search Central). That is verified platform guidance. It is not evidence that any compliant page will be selected or cited.

Audit technical foundations with reproducible evidence

Crawl the site and compare discovered URLs with sitemaps, analytics landing pages, Search Console data, and known business inventories. Group findings by template and rule. Check status codes, redirects, robots.txt, meta robots, X-Robots-Tag, canonicals, internal links, sitemap membership, mobile behavior, structured data, and rendering.

Keep control types distinct. Google documents that robots.txt manages crawling and does not guarantee removal from search results (Google Search Central). A canonical is a signal for preferred URL selection, not a redirect. A sitemap helps discovery and canonical signaling, but inclusion does not guarantee indexing.

Compare raw and rendered content on representative templates and exceptional states. Confirm that primary copy, links, titles, canonicals, directives, and JSON-LD survive rendering. Review Core Web Vitals with field data where available and use lab diagnostics to investigate components; do not mix both into one unlabeled score.

Leaf’s technical SEO audit checklist explains what evidence and closure criteria to require from a provider. The ten-step website SEO audit provides an operator workflow for running the checks.

Review content at both page and claim level

Map each priority page to a buyer, task, and commercial role. Check title and heading alignment, directness, depth, original value, source quality, update responsibility, and conversion path. Identify duplication and cannibalization based on actual intent, not merely shared words.

Then create a claim inventory for high-materiality facts: capabilities, limits, pricing model, integrations, certifications, deployment, customer eligibility, company identity, and evidence claims. Name the canonical maintained source and compare variants across product pages, documentation, PDFs, news, partner listings, and relevant external profiles.

Google’s helpful content guidance encourages original information, substantial coverage, clear sourcing, and people-first value. Use it as an editorial reference, not a ranking formula.

A good recommendation says what information is missing and why. “Add more content” is not useful. “Document the supported deployment models, constraints, owner, and last verification date on the canonical architecture page” is. Leaf’s content audit for AI search includes a practical claim-register format.

Check structured data and entity consistency

Inventory JSON-LD by template. Parse scripts independently, validate types and properties against Schema.org, and compare every material value with visible content. Use stable identifiers for the organization, products, services, and people where appropriate.

Google says valid structured data can make pages eligible for supported search features but does not guarantee display (Google Search Central). Its AI features guidance also says no special schema markup is required. Therefore, “add schema to guarantee AI citations” is not a defensible recommendation.

Use schema to describe real entities and relationships, not to publish hidden claims. Audit sameAs identity links, authors, dates, offers, ratings, and FAQ content especially carefully. See Leaf’s schema markup for AI search for a four-layer validation process and decision table.

Beyond markup, compare the company name, product taxonomy, executive identity, and category definitions across the site. Machines are not the only audience harmed by inconsistent naming; buyers and sales teams are too.

Sample answer visibility with a controlled prompt panel

Build a fixed panel from real buyer jobs: problem discovery, category education, solution comparison, implementation, objections, and vendor selection. Keep neutral discovery prompts separate from branded factual checks. Record exact wording and version the panel.

For each run, capture product, visible model or mode, account state, market, timestamp, prompt, full response, cited URLs, brand context, competitors, and material inaccuracies. Repeat high-value prompts because outputs vary.

Open citations and identify the claim each source appears to support. A brand mention is not a citation. A citation is not automatically an endorsement. A sampled mention order is not a universal rank. OpenAI’s current crawler documentation can inform access policy, but allowing a crawler does not prove retrieval or citation.

Leaf’s AI visibility audit describes evidence capture, while the guide to measuring AI search visibility shows how to report transparent denominators. Keep these observations separate from deterministic site checks.

Verify conversion and attribution measurement

Test analytics and CRM handoffs for commercially important actions. Confirm consent behavior, event names, deduplication, parameters, cross-domain paths, form receipt, and channel definitions. A dashboard configured months ago is not evidence that the current funnel records cleanly.

Track identifiable referrals from AI products, landing pages, engagement, qualified events, and self-reported discovery where the question can be asked neutrally. Referral information may be incomplete or grouped differently by analytics systems. State that limitation.

Do not turn citation coverage into a revenue estimate with an unsupported multiplier. Report three levels separately:

• Site controls: crawl, rendering, content, schema, and tracking checks.
• External observations: rankings, generated mentions, citations, and claim accuracy.
• Business outcomes: qualified sessions, conversions, opportunities, and revenue under the organization’s attribution rules.

A change in all three after a release is encouraging correlation. It is not automatically proof that one content or schema change caused the outcome.

Prioritize defects and experiments differently

Some findings have deterministic closure. A page returns 500, a canonical points to the wrong host, or a form event fires twice. Other recommendations are experiments: rewriting a comparison section, publishing primary research, or changing source presentation to test citation coverage.

Use this decision table:

Rank work by business exposure, severity, confidence, effort, and dependency. Give each item an owner. A high-volume list of warnings is not prioritization.

Use a 30-day implementation sequence

A realistic first cycle might look like this:

Week 1: establish the baseline. Freeze crawl settings and prompt panel, export evidence, validate conversion tracking, and confirm the revenue-page inventory.

Week 2: resolve foundational defects. Address access, response, canonical, rendering, sitemap, and critical measurement issues on important templates. Retest exact URL sets.

Week 3: repair information quality. Correct high-risk claims, clarify page purpose, improve primary sourcing, consolidate contradictions, and align entity details and markup.

Week 4: verify and observe. Re-crawl changed templates, test events, rerun the same prompt panel, compare source-level outcomes, and document what remains uncertain.

Do not force every issue into 30 days. Complex platform changes and field-performance measurement may need longer. The point is dependency order and a repeatable verification cycle.

What the final audit should contain

A finished combined audit should provide:

• scope, systems, access, methodology, dates, and limitations;
• critical page and template inventory;
• crawl and rendering evidence with affected URL exports;
• content and claim inventory for priority topics;
• structured-data and entity findings;
• versioned prompt panel and response evidence;
• analytics and conversion validation;
• prioritized issue register with owners and dependencies;
• deterministic retests and separately labeled monitored outcomes;
• executive summary focused on decisions, not warning counts.

If you want a bounded starting point, Leaf’s free AEO assessment checks observable homepage signals; it is not a full audit. Leaf’s SEO and AEO audit service covers the broader evidence and handoff for established B2B sites.

Choose the next investigation

Use the finding, not the fashionable acronym, to decide where to go deeper:

• For a software site whose product, comparison, and pricing pages carry most of the pipeline risk, use the B2B SaaS SEO and AEO audit protocol.
• When expertise, location, credentials, and lead quality matter more than product-led conversion, use the professional-services audit framework.
• If you are comparing providers, apply the AEO agency buyer questions before accepting a proprietary score or citation guarantee.
• If the team needs a bounded diagnostic before commissioning deeper work, compare the limits of free AEO audit tools.
• When terminology is creating more meetings than insight, settle the operating boundary with GEO vs AEO.
• If someone proposes llms.txt as a shortcut, review what llms.txt can and cannot do before assigning implementation time.
• For Google-specific visibility questions, use the documented controls in the AI Overviews guide rather than reverse-engineered guarantees.

Treat the audit as a baseline, not a verdict

Search systems, answer products, sites, and buyer language change. Preserve evidence, annotate releases, and retest important controls after implementation. Review material claims on a maintenance schedule and rerun the fixed prompt panel at a cadence the team can sustain.

The combined discipline is straightforward: prove access before optimizing extraction, establish source quality before chasing citations, and verify measurement before declaring impact. Your team can control the quality and availability of its information. Search and answer platforms still control indexing, ranking, retrieval, generation, and citation. A credible audit makes that boundary visible while giving operators concrete work to do.